Node Doctor checks the health of every Kubernetes node and can fix some problems on its own. It runs as a DaemonSet, writes what it finds as node conditions and events, and exports Prometheus metrics. It is modeled on Node Problem Detector and adds remediation.
- System: CPU load, memory and swap, OOM kills, disk space and inodes, read-only filesystems, and a disk write probe
- Network: DNS, default gateway, node-to-node connectivity over the CNI, IP forwarding, and IPv6 setup
- Kubernetes: kubelet health, API server latency, container runtime and pod capacity
- Custom: your own scripts, and regex matches on logs
Each finding becomes a node condition such as NodeDoctorKubeletUnhealthy. Remediation is opt-in per monitor and guarded by cooldowns, attempt limits, rate limits, a circuit breaker and dry-run mode. The optional controller collects reports from all nodes and limits how many nodes remediate at once.
helm repo add supporttools https://charts.support.toolshelm repo updatehelm install node-doctor supporttools/node-doctor --namespace node-doctor --create-namespaceThe image is supporttools/node-doctor on Docker Hub, for linux/amd64 and linux/arm64. Tags have no v prefix: release v1.10.0 is image 1.10.0.
Next, follow Quick Start to check that it works.
- Quick Start: install and check that it works
- Helm chart: every chart value
- Raw manifests: install without Helm
- Controller: the optional cluster-wide controller
- Dashboards: Grafana dashboards
- Troubleshooting: debug the agent or a node
- All docs: configuration, monitors, remediation, architecture, development
See CONTRIBUTING.md. Report security issues as described in SECURITY.md.
Apache License 2.0. See LICENSE.